It was held in Washington, D.C., last June 4-7, 2024. the Macro-XRF Scanning (MA-XRF) and Reflectance Imaging Spectroscopy (RIS) in Art, Conservation and Archeology 2024 conference (https://maxrf-ris-meeting-2024.academic.wlu.edu/) organized by the School of Engineering and Applied Science of George Washington University and the National Gallery in Washington. The conference, now in its third edition, was born as a meeting of experts focused on developing and using the non-invasive MA-XRF imaging technique in the study of tangible cultural heritage. In the 2024 edition, after the editions in Catania (Italy, 2019) and Delft (The Netherlands, 2022), the conference also included the reflectance imaging technique (RIS), the two analytical imaging methodologies today considered the cornerstones in the field of Heritage Science for the study of works of art and objects of historical-artistic interest.
CNR ISPC (the Institute of Heritage Science of the Italian National Research Council), which coordinates the Italian node (https://www.e-rihs.it/) of E-RIHS, the European Research Infrastructure for Heritage Science presented two case studies and new approaches in data analysis through machine learning, deep learning and artificial intelligence.
The case study on the Tonindeye Code
David Buti, from the CNR ISPC headquarters in Florence, presented some aspects of his research carried out in collaboration with the British Museum (https://www.britishmuseum.org/) and with the Santo Domingo Center of Excellence for Latin American Research (https ://www.sdcelarbritishmuseum.org/) as part of the short-term mobility 2023 program promoted by the CNR and supported by E-RIHS. During this period, some non-invasive investigation campaigns were carried out on a 25th-century Mesoamerican pictorial manuscript of Mixtec culture, the Tonindeye Codex, also known as the Zouche-Nuttall Codex. Thanks to the non-invasive approach and the services offered by the E-RIHS research infrastructure, it was possible to investigate the material composition of the pictorial materials used on the Mixtec code and investigate the creative process. The RIS technique has highlighted the use by Mesoamerican scribes for writing the code of different types of Mayan blue, a hybrid pigment composed of indigo and clay (palygorskite and/or sepiolite), considered the first nanocomposite material in history. The same technique not only revealed the production procedures of the Maya blue but also highlighted its specific use on the codex’s pages, giving scholars important information regarding its genesis, history and relationship with the other Mesoamerican codes.
The case study on the frieze of the tomb of King Philip II in Greece
Michela Botticelli, from the XRAYLab of the CNR ISPC branch in Catania, illustrated the results of the MOLAB (Mobile Laboratories platform of the E-RIHS infrastructure) campaign on the frieze of the monumental Macedonian tomb of King Philip II, father of Alexander the Great, at the Vergina site in Greece. The campaign was financed by the European project IPERION HS (Integrating Platforms for the European Research Infrastructure ON Heritage Science).
The case study presented is part of the REVIS project funded by the Hellenic Foundation for Research and Innovation in Greece, with Hariclia Brecoulaki of the Institute of Historical Research, National Hellenic Research Foundation in Athens as the Principal Investigator. The project sees the collaboration in Italy between two CNR Institutes, ISPC and SCITEC, and the University of Perugia with the SMAArt Center of Excellence; in Greece between the Institute of Nuclear and Particle Physics of the NCSR Demokritos, the Superintendency of Emanthia of the Greek Ministry of Culture, and the ISAAC Laboratory of the Nottingham Trent University in the UK.
The MOLAB access allowed the application of several cutting-edge analytical techniques, including MA-XRF, MA-XRD, HSI and RSI, which offered more in-depth knowledge of the original materials used by Macedonian artists. It was possible to demonstrate a much greater complexity of execution than that hypothesized, using both pigments common at the time (ochres, Egyptian blue) and others less known of probably local origin (copper arsenates and zinc). At the same time, the investigation highlighted iconographic details no longer visible to the naked eye, due to a compromised state of conservation. Various complementary techniques documented the presence of ‘non-original’ compounds attributable to degradation processes probably induced by the presence of microorganisms or particular environmental conditions. Finally, organic and inorganic compounds were found, which can be traced back to a previous conservative intervention.
During the activity, two X-ray techniques were used simultaneously (MA-XRF and MA-XRD) | © CNR ISPC
New machine learning, deep learning and artificial intelligence approaches for the analysis of spectral data
Zdenek Preisler, from the XRAYLab laboratory of the CNR ISPC branch in Catania, showed progress in machine learning and deep learning approaches applied to spectral data, particularly X-ray fluorescence data.
Current progress in non-invasive imaging methods applied to the study and conservation of cultural heritage has led to the rapid development of new computational methods. Still, even if the MA-XRF technique is well established and used for studying paintings, it generates large datasets that can be difficult to analyse. Machine learning approaches enable the identification of non-trivial dependencies and classifications among high-dimensional data, thus enabling comprehensive queries. A new deep learning algorithm trained on a synthetic MA-XRF dataset was presented at the conference, enabling fast and accurate analysis of XRF spectra while circumventing the typical drawbacks of the classical approach. Synthetic spectra are generated using “Monte Carlo” simulations based on fundamental parameters and optimized for the XRAYLab ‘s MA-XRF configuration. The simulations assume a stratigraphic model of a painting with many possible historical and modern pigments. The presented approach produces high-quality results in analysing MA-XRF scans and allows for methodology extensions and advanced applications.
The methodology was extended to paintings featuring changes in the original drawing (repentances) by modifying the neural network to include additional parameters relevant to more complex composition and stratigraphy, thus separating the underpaintings from the visible pictorial composition. In another application, a new artificial intelligence system was introduced, an AI/ML (artificial intelligence/machine learning) scheme that considers nearby spectra to increase the detection limit, allowing the image quality of less obvious elements to be improved. The examples demonstrate that these techniques open new scenarios and offer greater possibilities than the traditional approaches.
